Transient analysis of functionally graded plates using extreme gradient boosting

Abstract

This paper is aimed at quickly predicting the dynamic behavior of functionally graded plates using nontraditional computational approaches consisting of artificial neural networks (ANN) and extreme gradient boosting (XGBoost). Through the use of ANN and XGBoost, the dynamic behavior of the plate can be directly predicted based on optimal mapping, which is found by learning the relationship between input and output data from a data set during the training process. A data set including 1000 data pairs (input and output) is generated by using a combination of isogeometric analysis (IGA) and third-order shear deformation plate theory through iterations. In this model, a power index that controls the plate’s material distribution is regarded as input, and output consists of 200 values of deflection versus time. In order to demonstrate the effectiveness of XGBoost in terms of accuracy and computational time, results obtained by the optimal XGBoost model are compared to those obtained by the optimal ANN model and IGA.

A light gradient boosting machine-based method for predicting the dynamic response of functionally graded plates

“…In the previous study by the author, the accuracy of dynamic analysis of isotropic square plates using IGA and TSDT was verified [7]. In the present study, LightGBM has been utilized instead of IGA to examine dynamic response of the Al2O3/Al square plate (as shown in Figure 1) with 20 ha  and the damping ratio of 0.05 quickly.…”

“…These 1,000 data pairs are splitted into two groups: the first group contains 900 pairs that are used in the training phase, and the other contains 100 pairs that are used in the testing phase. Optimal models for ANN and XGBoost were determined after investigating the influence of various parameters on model accuracy and computational time presented in the authors' preceding study [7]. The aim of this study is to select the optimal LightGBM model by examining the impact of parameters such as the number of trees (n_estimators), maximum tree depth (max_depth), and learning rate (learning_rate) on the effectiveness of LightGBM.…”

“…Table 4 shows the effect of the learning rate on the accuracy and computational time of the LightGBM model. The parameters in the two models, ANN and XGBoost, were also examined in previous studies [7] to determine which model architecture was most effective in predicting the dynamic response of FG plates. Two optimal models were identified: XGBoost with max_depth of 5, n_estimators of 200, and a learning rate of 0.1; and the ANN model with Adam optimizer, ReLU function, 3,000 epochs, two hidden layers, and 50 nodes at each.…”

“…Applied in diverse engineering fields, IGA has been demonstrated for analyzing functionally graded plates [17], efficient shell analysis of complex multi-patch structures [18], and others [19][20][21][22][23]. The precision of isogeometric analysis (IGA) has been verified through a comparison of results with existing literature [7]. In this study, IGA is employed to generate a dataset comprising 1,000 pairs, each of which consists of a power index controlling the plate's material distribution as input and output with 200 deflection values over time.…”

“…The obtained results were compared with benchmark solutions, and the effects of parameters such as power index, porosity parameter, and side-to-thickness ratio on the transient response of FGM plates subjected to step blast loads were examined [6]. Artificial neural networks (ANN) and https://doi.org/10.55401/4x91s337 extreme gradient boosting (XGBoost) were utilized to rapidly predict the dynamic behavior of functionally graded plates [7]. Additionally, other investigations are documented in references [8,9].…”